Skip to content

zzf/reimpl slice with aclnn #1225

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: feat/replace_opplugin_by_aclnn
Choose a base branch
from
Open

zzf/reimpl slice with aclnn #1225

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ebe332d
reimpl slice with aclnn
zhangzefeng92 May 22, 2024
201f9aa
reimpl slice with aclnn
zhangzefeng92 May 22, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 1 addition & 9 deletions impl/ascend/aclnn/adaptor.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,15 +90,7 @@ inline aclTensor* createAclTensorFromDiopiTensor(diopiConstTensorHandle_t tensor
auto type = diopiDtypeToAclDataType(dtype);
auto format = inferAclDataFormat(shape.len, shape.data, stride.data);
auto storageSize = static_cast<int64_t>(storageNbytes / elemsize);
return ::aclCreateTensor(shape.data,
shape.len,
type,
stride.data,
storageOffset,
format,
&storageSize,
/*storageDimsNum=*/1,
const_cast<void*>(tensorData));
return ::aclCreateTensor(shape.data, shape.len, type, stride.data, storageOffset, format, shape.data, shape.len, const_cast<void*>(tensorData));
}

inline aclScalar* createAclScalarFromDiopiScalar(const diopiScalar_t* scalar) {
Expand Down
3 changes: 3 additions & 0 deletions impl/ascend/convert_config.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -444,3 +444,6 @@
- diopiLerpScalar:
# aclnnLerps doesn't support float64
dtype: (float64)->float32

- diopiSlice:
dtype: (float64)->float32, (int16)->int32
49 changes: 19 additions & 30 deletions impl/ascend/functions/slice.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,47 +4,36 @@
* @copyright (c) 2023, DeepLink.
*/

#include "../aclnn/adaptor.hpp"
#include "../common/acloprunner.hpp"

namespace impl {
namespace ascend {
diopiError_t diopiSlice(diopiContextHandle_t ctx, diopiTensorHandle_t nullOut, diopiConstTensorHandle_t input, int64_t dim, int64_t start, int64_t end,
int64_t step) {
std::vector<int64_t> index;
for (int64_t i = start; i < end; i += step) {
index.push_back(i);
}
std::vector<int64_t> dimVec{dim};
bool negativeIndexSupport = true;
AclOpRunner<3, 1>("GatherV2", ctx)
.addInput(input)
.addConstInput(vectorToDiopiSize(index))
.addConstInput(vectorToDiopiSize(dimVec))
.setAttr("negative_index_support", negativeIndexSupport)
.addOutput(nullOut)
.run();
DIOPI_ASCEND_CALL_ACLNN(aclnnSlice, ctx, input, dim, start, end, step, nullOut);
return diopiSuccess;
}

diopiError_t diopiSliceBackward(diopiContextHandle_t ctx, diopiTensorHandle_t gradInput, diopiConstTensorHandle_t gradOutput, diopiSize_t inputSizes,
int64_t dim, int64_t start, int64_t end, int64_t step) {
if (dim < 0) {
dim += inputSizes.len;
}
std::vector<int64_t> start64(inputSizes.len, 0);
std::vector<int64_t> step64(inputSizes.len, 1);
std::vector<int64_t> end64(inputSizes.data, inputSizes.data + inputSizes.len);
start64[dim] = start;
step64[dim] = step;
end64[dim] = end;
AclOpRunner<5, 1>("StridedSliceGrad", ctx)
.addConstInput(inputSizes)
.addConstInput(vectorToDiopiSize(start64))
.addConstInput(vectorToDiopiSize(end64))
.addConstInput(vectorToDiopiSize(step64))
.addInput(gradOutput)
.addOutput(gradInput)
.run();
DIOPI_ASCEND_CALL_ACLNN(aclnnInplaceZero, ctx, gradInput);

if (start < 0) start += inputSizes.data[dim];
if (end < 0) end += inputSizes.data[dim];

diopiTensorHandle_t index;
std::array<int64_t, 1> indexSize = {(end - start) / step};
diopiSize_t indexShape = {indexSize.data(), 1};
diopiRequireTensor(ctx, &index, &indexShape, nullptr, diopi_dtype_int64, diopi_device);

diopiScalar_t startScalar = constructDiopiScalarT(diopi_dtype_int64, start);
diopiScalar_t endScalar = constructDiopiScalarT(diopi_dtype_int64, end);
diopiScalar_t stepScalar = constructDiopiScalarT(diopi_dtype_int64, step);
DIOPI_ASCEND_CALL_ACLNN(aclnnArange, ctx, &startScalar, &endScalar, &stepScalar, index);

diopiScalar_t one = constructDiopiScalarT(diopi_dtype_int64, 1);
DIOPI_ASCEND_CALL_ACLNN(aclnnIndexAdd, ctx, gradInput, dim, index, gradOutput, &one, gradInput);

return diopiSuccess;
}
Expand Down